Pietro Lesci (@pietro_lesci) 's Twitter Profile
Pietro Lesci

@pietro_lesci

PhD student @cambridge_uni. Causality & language models. Former @bainandcompany, intern @ecb @amazonscience. Passionate musician, professional debugger.

ID: 1018592521600028679

linkhttps://pietrolesci.github.io calendar_today15-07-2018 20:25:56

725 Tweet

642 Followers

1,1K Following

EleutherAI (@aieleuther) 's Twitter Profile Photo

Like Pythia, but 1234 isn't your favorite random seed? We retrained Pythia 9x using different random seeds to explore how stable analyses of learning dynamics are to randomness. Meet Pietro Lesci @blancheminerva Fri 1500-1730 Hall 3 + Hall 2B #259 x.com/pietro_lesci/s…

Nedjma Ousidhoum نجمة أوسيدهم (@nedjmaou) 's Twitter Profile Photo

Happy to see our paper win ✨the Best Theme Paper Award✨ at #NAACL2025 ! Working on this project with this team was a lot of fun :-) huge thanks to Genta Winata, Frederikus Hudi,Patrick Amadeus,@davidanugraha, Rifki Afina Putri for leading!

Andreas Vlachos (@vlachos_nlp) 's Twitter Profile Photo

The call for papers for the 8th FEVERworkshop at #ACL is out: fever.ai/workshop.html Deadline for is on May 19th! And if you have a paper already reviewed in ARR, you can commit it until June 9th!

Workshop on Large Language Model Memorization (@l2m2_workshop) 's Twitter Profile Photo

📢 ACL 2025 notifications have been sent out, making this the perfect time to finalize your commitment. Don't miss the opportunity to be part of the workshop! 🔗 Commit here: openreview.net/group?id=aclwe… 🗓️ Deadline: May 20, 2025 (AoE) #ACL2025 #NLProc

Tiago Pimentel (@tpimentelms) 's Twitter Profile Photo

If you're finishing your camera-ready for ACL (#acl2025nlp) or ICML (#icml2025 ) and want to cite co-first authors more fairly, I just made a simple fix to do this! Just add $^*$ to the authors' names in your bibtex, and the citations should change :) github.com/tpimentelms/ac…

If you're finishing your camera-ready for ACL (#acl2025nlp) or ICML (#icml2025 ) and want to cite co-first authors more fairly, I just made a simple fix to do this! Just add $^*$ to the authors' names in your bibtex, and the citations should change :)

github.com/tpimentelms/ac…
Tiago Pimentel (@tpimentelms) 's Twitter Profile Photo

A string may get 17 times less probability if tokenised as two symbols (e.g., ⟨he, llo⟩) than as one (e.g., ⟨hello⟩)—by an LM trained from scratch in each situation! Our #acl2025nlp paper proposes an observational method to estimate this causal effect! Longer thread soon!

A string may get 17 times less probability if tokenised as two symbols (e.g., ⟨he, llo⟩) than as one (e.g., ⟨hello⟩)—by an LM trained from scratch in each situation! Our #acl2025nlp paper proposes an observational method to estimate this causal effect! Longer thread soon!
Tiago Pimentel (@tpimentelms) 's Twitter Profile Photo

If you use LLMs, tokenisation bias probably affects you: * Text generation: tokenisation bias ⇒ length bias 🤯 * Psycholinguistics: tokenisation bias ⇒ systematically biased surprisal estimates 🫠 * Interpretability: tokenisation bias ⇒ biased logits 🤔